Practical guidelines for handling head and neck computed tomography artifacts for quantitative image analysis
Introduction
Radiomics, the analysis of medical images on a voxel level to extract quantitative image features, has become a popular research technique in oncology. Radiomics is based on the assumption that the gene microenvironment is expressed on a macroscopic level and that this information can be extracted by analyzing voxel-level data in various ways (Lambin et al., 2012). Therefore, by extracting texture metrics from the image, information inaccessible to the human eye alone can be obtained. This additional information has been shown, for instance, to improve the ability of survival models to distinguish patients by prognosis when added to conventional prognostic factors such as age (Bogowicz et al., 2017a,b; Fave et al., 2017; Fried et al., 2016; Ou et al., 2017; Vallieres et al., 2017). Most early radiomics studies focused on lung cancer, but patients with head and neck cancer have recently become a prominent focus of radiomics studies (Bagher-Ebadian et al., 2017; Bogowicz et al., 2017a,b; Ou et al., 2017; Parmar et al., 2015b,a; Vallieres et al., 2017).
While radiomics studies have identified several imaging features that are associated with prognosis, these findings can be affected by a variety of factors. The impact of many characteristics of imaging protocols, such as voxel size, tube current, tube voltage, and kernel, has been studied thoroughly (Fave et al., 2015; Mackin et al., 2017, 2018; Shafiq-ul-Hassan et al., 2017; Zhao et al., 2014, 2016). However, the effects of factors intrinsic to the patient have not been investigated. For example, computed tomography (CT) scans of the head and neck cover the oral cavity, where many patients have metal dental fillings that cause streak artifacts. As radiomics is based on the assumption that gene expression at a microscopic level is discernible on a macroscopic level in the voxels, it is likely that measuring the radiomics features of the structures affected by a streak artifact would not provide any valuable information about that structure. Another type of artifact observed in CT scans, beam hardening, can affect images containing bone. Because there are many bones in the area of interest in head and neck examinations, this area may be particularly prone to the effects of these small artifacts. As a result, patients whose structure of interest is affected by streak or beam-hardening artifacts are often excluded from the large data sets required to achieve sufficient statistical power for radiomics studies. Therefore, finding a way to include as many patients as possible is needed.
We aimed to test the impact of these artifacts and if needed, methods for compensating for these artifacts in head and neck radiomics studies. First, we determined whether streak artifacts do in fact alter radiomics feature values, and, if so, whether the simple technique of removing the slices affected by the streak artifact produced feature values similar to those in regions unaffected by the artifact. Second, we aimed to determine whether a buffer region is needed between bone and other structures to ensure that the measured feature values are not affected by beam-hardening artifacts.
Section snippets
Impact of streak artifacts on feature values
The impact of streak artifacts on feature values was investigated using a cohort of 458 patients with head and neck squamous cell carcinoma (HNSCC). All procedures were performed in accordance with the Declaration of Helsinki on Ethical Issues with a waiver of informed consent from the Institutional Review Board at the University of Texas MD Anderson Cancer Center. Only the patients whose CT images exhibited a visible streak artifact on slices showing the gross tumor volume (GTV) were selected,
Impact of streak artifacts on feature values
On average, 3.0 cm3 of GTV had to be removed to eliminate streak artifacts (standard deviation: 4.0 cm3, range: 0.11–28 cm3). Table 1 shows the percentage of features for which the measured value in the original GTV (with artifact) and the modified GTV (without artifact) differed significantly. Only for gray-level run length matrix features preprocessed using thresholding and intensity features preprocessed using thresholding, smoothing, and 8-bit depth resampling were fewer than 70% of the
Discussion
In this study, we showed that streak artifacts affect radiomics feature values, suggesting that regions containing such artifacts should not be included in radiomics data sets. We demonstrated that a simple technique, removing the slices with the artifact, can be used to remove up to 50% of the original GTV from the ROI while retaining similar feature values. Additionally, while the presence of bone within the image can affect some feature values, the effect is typically smaller than the spread
Conclusion
We demonstrated that streak artifacts affect the measured radiomics feature values. In order to deal with this effect, we suggest simply removing the slices with the artifact. Using this method, feature values are robust when up to 50% of the original GTV is removed. Excluding patients in whom more than 50% of the GTV is affected by the artifact only causes about 3% of patients to be excluded. Additionally, we demonstrated that contours can abut bone if needed, as most features are not affected
Declarations of interest
None.
Acknowledgements
This work was supported by the National Institutes of Health [grant #: R21CA216572]. Rachel Ger is supported by the Rosalie B. Hite Graduate Fellowship in Cancer Research and the American Legion Auxiliary Fellowship in Cancer Research awarded by the MD Anderson Cancer Center UTHealth Science Center at Houston Graduate School of Biomedical Sciences. The authors would like to acknowledge the editing assistance of the Department of Scientific Publications at MD Anderson Cancer Center.
References (28)
Computed tomography radiomics predicts HPV status and local tumor control after definitive radiochemotherapy in head and neck squamous cell carcinoma
Int. J. Radiat. Oncol. Biol. Phys.
(2017)Preliminary investigation into sources of uncertainty in quantitative imaging features
Comput. Med. Imaging Graph.
(2015)Texture analysis using gray level run lengths
Comp. Graph. Image Process.
(1975)Radiomics: extracting more information from medical images using advanced feature analysis
Eur. J. Cancer
(2012)Predictive and prognostic value of CT based radiomics signature in locally advanced head and neck cancers patients treated with concurrent chemoradiotherapy or bioradiotherapy and its added value to human papillomavirus status
Oral Oncol.
(2017)Exploring variability in CT characterization of tumors: a preliminary phantom study
Transl. Oncol.
(2014)- et al.
Textural features corresponding to textural properties
IEEE Trans. Syst. Man. Cybern.
(1989) On the impact of smoothing and noise on robustness of CT and CBCT radiomics features for patients with head and neck cancers
Med. Phys.
(2017)Comparison of PET and CT radiomics for prediction of local tumor control in head and neck squamous cell carcinoma
Acta Oncol.
(2017)- et al.
Preparation and fabrication of a full-scale, sagittal-sliced, 3D-printed, patient-specific radiotherapy phantom
J. Appl. Clin. Med. Phys.
(2017)
Material matters: analysis of density uncertainty in 3D printing and its consequences for radiation oncology
Med. Phys.
Impact of image preprocessing on the volume dependence and prognostic potential of radiomics features in non-small cell lung cancer
Transl. Cancer Res.
Delta-radiomics features for the prediction of patient outcomes in non-small cell lung cancer
Sci. Rep.
Stage III non-small cell lung cancer: prognostic value of FDG PET quantitative imaging features combined with clinical prognostic factors
Radiology
Cited by (27)
The Promise and Future of Radiomics for Personalized Radiotherapy Dosing and Adaptation
2023, Seminars in Radiation OncologyRadiotherapy outcome prediction with medical imaging
2023, Machine Learning and Artificial Intelligence in Radiation Oncology: A Guide for CliniciansDiagnostic Accuracy and Failure Mode Analysis of a Deep Learning Algorithm for the Detection of Intracranial Hemorrhage
2021, Journal of the American College of RadiologyDevelopment and dosimetric assessment of an automatic dental artifact classification tool to guide artifact management techniques in a fully automated treatment planning workflow
2021, Computerized Medical Imaging and GraphicsAutomated detection of dental artifacts for large-scale radiomic analysis in radiation oncology
2021, Physics and Imaging in Radiation OncologyCitation Excerpt :For these features to be predictive of radiotherapy response, they must be highly reproducible and safeguards for data corruption must be put in place [3]. Unfortunately, radiomic features may be highly sensitive to high-density materials such as metal prosthesis or dental fillings [4]; the latter commonly causes dental artifacts, which pose a problem for imaging of head and neck patients. The metal in dental fillings has a much larger atomic number than soft tissues, resulting in a significantly higher attenuation for x-ray beams passing through the metal.
Prediction of post-radiotherapy locoregional progression in HPV-associated oropharyngeal squamous cell carcinoma using machine-learning analysis of baseline PET/CT radiomics
2021, Translational OncologyCitation Excerpt :PET segmentations were then copied onto the co-registered CT and manually adjusted to the GTV outline on CT using the “Paint” and “Erase” tools to generate the CT VOI, excluding air, adjacent uninvolved bone, and preserved fat planes. Axial CT slices with streak artifacts involving the lesion upon visual assessment were excluded from analysis; and metastatic lymph nodes with >50% of the GTV involved were entirely excluded [25]. A trained research associate (SPH) initially segmented all lesions; followed by VOI verification and adaption by a neuroradiologist (SP) with greater than 8 years of experience in head and neck cancer imaging.